Ambient context in breathing signal fusion

Fusion of ambient breathing signals can be hampered by the current environmental conditions. Fusion methods that are resistant to most adverse ambient conditions may still be susceptible to some. The diversity of multiple fusion methods can be leveraged by employing a variety of fusion methods to fuse each breathing epoch. By using a trained linear classifier to select the best candidate fusion method according to current condition estimates, a more robust system is proposed. Regularization by feature selection and method selection were addressed. The final system used two fusion methods and ten ambient features to outperform the single best fusion method. The system increased the percentage of epochs with good signal quality by 5%. Analysis of optimal fusion method selection indicates that further improvement is also possible.